New Approach on Analysis of Pathologic Cardiac Murmurs...

Journal of Healthcare Engineering · Vol. 5 · No. 4 · 2013 Page 375–392 375

New Approach on Analysis of Pathologic CardiacMurmurs Based on WPD Energy Distribution

Zhongwei Jiang1*, Ting Tao1 and Haibin Wang2

1Graduate School of Science and Engineering, Yamaguchi University, Yamaguchi,Japan,

2School of Electrical and Information Engineering, Xihua University, Chengdu, China

Submitted March 2014. Accepted for publication September 2014.

ABSTRACTIn this paper, an approach on analysis of the pathologic cardiac murmurs for congenital heartdefects was proposed based on the wavelet packet (WP) technique. Considering the difference ofthe energy intensity distributions for the innocent and pathologic murmurs in frequency domain,the WP decomposition was introduced and the WP energies at each frequency band werecalculated and compared. Based on the analysis of a large amount of clinic heart sound data, themurmurs energy distributions were divided into five frequency bands, and the relative evaluationindexes for cardiac murmurs (ICM) were proposed for analysis of the pathologic murmurs.Finally, the threshold values between the innocent and pathologic cardiac murmurs weredetermined based on the statistical results of the normal heart sounds. The analysis resultsvalidate the proposed evaluation indexes and the corresponding thresholds.

Keywords: cardiac murmurs analysis, energy distribution, wavelet decomposition, evaluationindex, congenital heart defects

1. INTRODUCTIONCongenital heart defects (CHDs) are problems with the heart’s structure or functionthat is formed before birth [1], with a rate of eight out of every 1,000 newborns. Itis known that this kind of defects can be cured with a high probability if the diseasescould be detected in an early stage. Therefore, the research on the early detection ofCHDs is one of the most important medical research areas [2]. There are many typesof congenital heart defects. Some are simple and some are more complex. Atrialseptal defect (ASD) is one of the CHDs, a hole in the part of the septum thatseparates the atria. The hole makes oxygen-rich blood from the left atrium to flowinto the right atrium, instead of flowing into the left ventricle as it should. ASDs canbe small, medium, or large. Medium and large ASDs allow more blood to leak from

*Corresponding author: Zhongwei Jiang, Tokiwadai 2-16, Ube, Yamaguchi, 755-8611, Japan. Phone: (81)836 859137. Fax: (81)836 85 9101. E-mail: [email protected]. Other authors: [email protected];[email protected].

one atrium to the other. Ventricular septal defect (VSD) is another type of CHD, with a hole in the part of the septum that separates the ventricles. Small VSDs do not cause problems and may close on their own. However, large VSDs allow muchblood to flow from the left ventricle to the right ventricle. As a result, the left sideof the heart must work harder than normal and extra blood flow increases bloodpressure in the right side of the heart and the lungs. Furthermore, the most commoncomplex heart defect is tetralogy of Fallot (TOF), which is a combination of four defects: Pulmonary valve stenosis, a large VSD, an overriding aorta and right ventricular hypertrophy. ASD, VSD and TOF account for the majority of the CHD [3].

Heart murmurs are pathological sounds produced by turbulent blood flow due tocertain cardiac defects, and they are the most common reasons for referral to thepediatric cardiologist. In children, about 50-70% of these murmurs are clinicallyinsignificant [4], but if the child is crying, uncooperative to the examiner or breathingloudly, some other murmurs may occur. Because of the difficulty of masteringauscultation skills, innocent and organic heart murmurs cannot be readilydistinguished. Therefore, heart murmur quantitative analysis is necessary. Recently,the demand for evaluation of the murmurs from auscultation of cardiac sounds hasbeen addressed by researchers and clinicians [5-8]. To detect and analyze the heartsound (HS) murmurs, many approaches have been carried out, including a multivariatematching pursuit method [7] used to model the murmurs by decomposing them into aseries of parametric time-frequency atoms, and then the model parameters used toidentify the cardiac sound signals. An adaptive singular spectrum analysis approach[8] was applied to the changes in the statistical properties of the sound data fordetection of murmurs.

Since the heart murmurs show clearly different characteristics in the frequencydomain compared with the time domain, many researchers have focused on thecharacteristic extraction by local frequency analysis method, such as the waveletdecomposition (WD) or WPD [9-14, 23], the neural network [6, 15-18], support vectormachine(SVM) [19-21]. Sepehri, et al. [18], studied a method for automatic screeningof CHD in children with neural network classifier. The pathological murmurs of CHDwere identified by examining the HS energy over special frequency bands called Arash-Bands. However, the Arash-Bands determined for each CHD in the study wereoverlapped with the frequencies 16-121 Hz. Based on our study, the main energydistributions either for normal or abnormal HSs are concentrated in the frequency rangeof 10-100Hz. We also found that frequency band is much influenced by measurementsituations, ages, body types, mixed heart defects, etc. We have tried many methods butit seems difficult to identify some murmurs by the Arash-Bands. Furthermore, SamjinChoi, et al. [9, 23] proposed insufficiency murmur identification and valvular disordersdetection using wavelet packet decomposition, and HS signals with frequency range 20-700Hz were preferred; furthermore, the features were extracted from WP coefficientcalculated for each subdivision. However, the subjects in these two studies were fromthe medical text book [26] or internet web [27]. It is not clear whether WP coefficient

376 New Approach on Analysis of Pathologic Cardiac Murmurs Based on WPD Energy Distribution

calculated for each subdivision with the frequency range 20-700Hz can identify theCHD signals in general auscultation environment in hospital.

In our study, a novel approach is investigated for analysis of the pathologic cardiacmurmurs of CHD with the energy distributions over special frequency bands. Theclinical CHD including the most common CHD of children (3 month to 10 years old,several are 24~27 years old) is investigated. In collecting the clinical signals fromchildren in hospital, we found that children’s thoracic cavities are very thin, andduring clinical auscultation, high frequency (344-1378Hz) HS components producedby blood flow through the valves are easy to collect through the thin chest, and theheart beat activity always cause chest vibration through the thin chest. Hence, thevery low frequency (5-21Hz) component from chest vibration or body movement canbe collected. Therefore, in our study, we preferred the heart signals with thefrequency range with 5-1378Hz. According to the analysis results, we also found thatfrequency division too small would highlight breathe sound component and othernoises from body moving or environment, which would influence the identificationresult. Therefore, in our study five frequency band divisions were employed; thewavelet packet decomposition (WPD) is introduced and the WP energy distributionsat these five frequency bands are calculated and compared. The evaluation indexesfor cardiac murmurs are proposed for the analysis of the pathologic murmurs. Finally,the threshold values between the innocent and pathologic murmurs are determinedbased on the statistical results of the HSs. The analysis results demonstrate thevalidation and efficiency by using the proposed evaluation indexes and thecorresponding thresholds.

2. METHODAuscultation denotes the act of analyzing sounds from the body, which is producedby the mechanical vibrations generated in the organs. The HSs are primarilygenerated from blood turbulence. Heart defects can cause heart murmurs (extra orunusual sounds heard during a heartbeat). Doctors can hear heart murmurs using astethoscope. In CHD patients, a pulmonary ejection click or/and soft and scratchysystolic ejection murmurs are audible [22]. HS auscultation is usually performed atfour locations (aortic, pulmonic, tricuspid and mitral areas) in front of the chest. Inthis paper, the cardiac sound signals auscultated from the pulmonic area wereselected as the analysis signals because the HS is most audible in this area. The use of the data in this study has been approved by local data collection ethicscommittee.

The cardiac murmurs for different heart defects usually contain differentfrequency components. If the sound signal can be decomposed in different requestedfrequency ranges in a suitable way, the corresponding energy intensities can beevaluated quantitatively. In this study, the cardiac signal is decomposed andreconstructed at each requested frequency band by the WPD technique and theirmurmurs energy intensities are calculated and used for the cardiac murmursevaluation. The block diagram for murmurs analysis is depicted in Figure 1 and thedetail is described below.

Journal of Healthcare Engineering · Vol. 5 · No. 4 · 2014 377

2.1. Pre-processingThe original HS signals were recorded by an electric stethoscope with 16-bit accuracyand 44.1 kHz sampling frequency. The recorded signal x(i) is first down-sampled to11025 Hz to reduce the calculation time. Next, WPD implementation is applied to thesignal with MATLAB software. Daubechies 10 (DB10) wavelet was used as the motherwavelet because it very well addresses biomedical signals [9, 20, 24]. The originalsignal x(i) is first preprocessed by the WPD-based filter with frequency band of 5.38-1378.1 Hz. The obtained signal xm(i) is then normalized by Eqn. (1) and used in thefollowing analysis:

(1)

2.2. Frequency Band DefinitionBy considering the frequency ranges of the innocent and pathologic murmurs in thefrequency domain, in this study, the wavelet packet decomposition at level 10 is employedto split frequency bandwidths of HS signals. Through clinical auscultation observationand energy intensities analysis at each level (d10~d3), the interested parts and usefulcombinations are implemented, and five frequency bands are defined and shown in Table1; they are named as very low frequency (VLF), standard frequency (SF), low frequency(LF), middle frequency (MF) and high frequency (HF). Their corresponding waveletdetail coefficient levels and center frequencies are also presented in Table 1.

( )=x i

x i

x iˆ( )

( )

max ( )m

m


NHSCHD-ASDCHD-VSDCHD-TOF

HSsignals

Pre-processingprocedure

Frequency banddefinition

Covert sampling rateWavelet filterNormalize HS

Select and definethe frequency bands

WPDimplementation

Energy distribution indifferent ranges

Murmurs evaluationwith ICM and TCM

Calculation ofWP energy ratio

Figure 1. The block diagram for murmurs analysis.

Table 1. Murmurs frequency bands definition.

Frequency band number (j) 1 2 3 4 5

Frequency band VLF SF LF MF HFWavelet detail coefficients d10+d9 d8+d7 d6 d5 d4+d3Frequency band corresponded to sampling frequency (Hz) 5.38-21.53 21.53-86.1 86.1-172.2 172.2-344.4 344.4-1378Center frequency (Hz) 7-14 29-59 117 235 471-942(Measured center frequency) (6-12.5) (25-50) (100) (200) (400-800)


2.3. WPD ImplementationThe wavelet packet analysis is an extension of the discrete wavelet transform (DWT)and is one of the most efficient decomposition that could be performed on the signal.Daubechies type wavelet (DB10) is used to build a wavelet filter. The centerfrequencies are calculated from the mother wavelet (DB10) at the corresponding detaillevels (Table 1). Further the reconstructed signals at each decomposition level areapplied by FFT and their peak frequencies are measured and named as measured centerfrequency (Table 1) just for reference. Figure 2 shows the plots of some typical HSstreated by WP analysis method. Figure 2(a)-(d) are the examples of NHS, ASD, VSDand TOF signals obtained by reconstruction of the components at VLF, SF, LF, MF andHF band as defined in Table 1. A simple sum of all the VLF, SF, LF, MF and HFcomponents represent the original HS signal.

The NHS signal plots at each frequency band show the 1st and 2nd sounds clearly andthe energy is mainly distributed at the SF band. As for ASD, the signal plots show thatthe murmurs appeared mainly at LF and MF bands. As for VSD and TOF, the murmurshave stronger energy distributions at MF and HF frequency bands.

2.4. WP Energy (WPE) RatiosFigure 2(a)-(d) show that the energy of all types of HS signals is mainly concentratedat SF band. Further, compared to NHS, CHD signals have higher energy intensities atthe higher frequency bands. To quantify the cardiac murmurs, WP energy ratio Êj isproposed and defined as the following:

(2)

x̂j (i) is the reconstructed signal at the corresponding frequency band j, N is the signallength and j = 1, 2, 3, 4, 5 are corresponding to frequency bands VLF, SF, LF, MF andHF respectively.

3. DATA ANALYSISThe original HS signals were recorded by an electric stethoscope with 16-bitaccuracy and 44.1 kHz sampling frequency. HS data were collected from 85 testsubjects who have signed an informed consent. The subjects included 21 healthyyoung college students and 64 CHD patients. CHDs patients included 17 ASD, 29VSD and 18 TOF.

The energy distributions for these four conditions (NHS, ASD, VSD and TOF) arecalculated at each frequency band defined in Table 1. The energy distributionhistograms at SF, LF, MF and HF frequency bands are plotted in Figure 3 to Figure 6.The y-axis shows the number of samples or frequency of the observations in the intervaland the x-axis is the energy ratio Êj in percentage. There are about 40% to 90% energydistributed at SF frequency band, 5% to 30% at LF band, 1% to 10% at MF and 0.1%to 4% at HF band.

∑ ∑= × = == =

EE

EE x i E Eˆ 100%, where ˆ ( ) andj

jj j

i

N

jj

n

total 1

2

total1


0 1 2 3−1

−0.50

0.51

Orig

inal

0 1 2 3−1

−0.50

0.51

VLF

0 1 2 3−1

−0.50

0.51

SF

0 1 2 3−1

−0.50

0.51

LF0 1 2 3

−0.5

0

0.5

MF

0 1 2 3−0.5

0

0.5

HF

(a) NHS

0 1 2 3−1

−0.50

0.51

Orig

inal

0 1 2 3−1

−0.50

0.51

VLF

0 1 2 3−1

−0.50

0.51

SF

0 1 2 3−1

−0.50

0.51

LF

0 1 2 3−0.5

0

0.5

MF

0 1 2 3−0.5

0

0.5

HF

(b) ASD

Figure 2. (Continued)


0 1 2 3−1

−0.50

0.51

Orig

inal

0 1 2 3−1

−0.50

0.51

VLF

0 1 2 3−1

−0.50

0.51

SF

0 1 2 3−1

−0.50

0.51

LF

0 1 2 3−0.5

0

0.5

MF

0 1 2 3−0.5

0

0.5

HF

(c) VSD

0 1 2 3−1

−0.50

0.51

Orig

inal

0 1 2 3−1

−0.50

0.51

VLF

0 1 2 3−1

−0.50

0.51

SF

0 1 2 3−1

−0.50

0.51

LF

0 1 2 3−0.5

0

0.5

MF

0 1 2 3−0.5

0

0.5

HF

(d) TOF

Figure 2. HS signal examples of NHS, ASD, VSD, TOF and their reconstructedsignals at different corresponding frequency band, Original, VLF, SF, LF,MF and HF.


0 5 10 15 20 25 30 350

10

20(a) NHS

0 5 10 15 20 25 30 350

5

10(b) ASD

0 5 10 15 20 25 30 350

5

10(c) VSD

Num

ber

of s

ampl

es

0 5 10 15 20 25 30 350

5(d) TOF

ELF (%)

Figure 4. Energy distribution at LF frequency.

30 40 50 60 70 80 90 100

30 40 50 60 70 80 90 100

30 40 50 60 70 80 90 100

30 40 50 60 70 80 90 100

0

5

10(a) NHS

0

2

4(b) ASD

0

5

10(c) VSD

Num

ber

of s

ampl

es

0

2

4(d) TOF

ESF (%)

Figure 3. Energy distribution at SF frequency.


0 1 2 3 4 5

0 1 2 3 4 5

0 1 2 3 4 5

0 1 2 3 4 5

0

10

20(a) NHS

0

5

10(b) ASD

0

2

4(c) VSD

Num

ber

of s

ampl

es

0

2

4(d) TOF

EHF (%)

Figure 6. Energy distribution at HF frequency.

0 5 10 150

10

20(a) NHS

0 5 10 150

5(b) ASD

0 5 10 150

5

10(c) VSD

Num

ber

of s

ampl

es

0 5 10 150

2

4(d) TOF

EMF (%)

Figure 5. Energy distribution at MF frequency.

The averages and the standard deviations for these four conditions are summarizedin Table 2. Statistical analysis was performed to show whether there is a significantdifference in means among these four conditions (NHS, ASD, VSD and TOF) using the

one-way ANOVA F-test. The very low P-values (<0.0001) were taken as statisticallysignificant. It is obvious that the distributions of NHS are following the normalprobability distribution. The energy distributions of ASD are similar to NHS.Furthermore, the energy distributions of other CHDs are spread over a wide range andthe energies maintain higher values at MF and HF bands compared to those in NHS.

Table 2 shows over 50% of energy in average concentrated at SF band for allconditions. This shows that the main component of HS is in the SF band. Further,Figure 2 and Table 2 indicate that the pathologic murmurs feature high energydistributions at LF, MF and HF.

On the other hand, the energy distribution at VLF band is considered to be avibration or systaltic movement in the body due to the heartbeat. The energydistribution histograms at VLF band are shown in Figure 7. For NHS, the energy is


0 10 20 30 40 50 60

0 10 20 30 40 50 60

0 10 20 30 40 50 60

0 10 20 30 40 50 60

0

5

10(a) NHS

0

2

4(b) ASD

0

5

10(c) VSD

Num

ber

of s

ampl

es

0

2

4(d) TOF

EVLF (%)

Figure 7. Energy distribution at VLF frequency.

Table 2. Means and variances of the energy ratios at four frequency bands (P < 0.0001)

Group(Eave±Estd) NHS ASD VSD TOF P-value

EVLF(%) 15.67±4.45 20.30±19.18 13.31±13.33 30.69±19.84 <0.0001ESF(%) 76.16±4.33 68.67±18.95 66.90±9.94 53.45±13.63 <0.0001ELF(%) 6.05±1.66 7.51±3.89 13.58±6.91 9.60±5.41 <0.0001EMF(%) 1.82±0.792 2.65±1.76 4.692±2.83 4.58±2.86 <0.0001EHF(%) 0.29±0.12 0.84±0.74 1.52±0.99 1.68±1.26 <0.0001


Figure 8. Distributions of the indexes of cardiac murmurs at LF band (ICMLF).

usually lower than 25% but for CHDs, the energies sometimes are higher than 30%. Itwas found that the HSs recorded from the infant usually contain high energycomponents at VLF band and some of them account for more than 40% of the totalenergy. Because the infant has little fat, the pulsation of the heart easily leads to amovement of the whole chest region. This movement, especially due to the abnormalheart pulsation is captured by another stethoscope which shows the high energycomponent at VLF band. Since there are still many uncertainties for this VLF band, thisstudy has mainly concentrated on the four bands, SF, LF, MF and HF.

4. RESULTS AND DISCUSSIONSAs discussed above, the main energy of HS is concentrated at SF band and thepathologic murmurs represent the high energy density at LF, MF and HF bands. Basedon this fact and our investigation, the evaluation indexes of cardiac murmurs areproposed and defined as the following:

(3)

Figure 8 shows the plots of ICMLF for NHS and CHDs. The average and standarddeviation of ICMLF for NHS are obtained as 8.01±2.44. The threshold for cardiacmurmurs (TCM) at LF band is defined by TCMLF = 8.01+2.44 = 10.45, depicted as abroken line in the figure. It means that it could be diagnosed to be the pathologicmurmurs if the obtained energy level is higher than this threshold TCMLF. From Figure8, it is found that nine of the 17 ASD samples are below the threshold TCMLF,indicating nine subjects in false diagnosis. On the other hand, only four of the 29 VSDand three of the 18 TOF samples are below the threshold.

Point in Figure 8 has the highest value of ICMLF in NHS at LF band, which isbeyond the threshold of TCMLF. This HS signal is further analysed and the results areplotted in Figure 9. It is confirmed that the strong energy component at LF band is thepulmonary sound due to a normal breath. The breath component can also be found at

{ }= × =ICME

EL LF MF HF

(%)

(%)100%, where , ,L

L

SF

A

10.45 A

0 10 20 300

10

20

30

40

50

Sample number0 10 20 30


Sample number

ELF

/ES

F (

%)

0

10

20

30

40

50

ELF

/ES

F (

%)

0

10

20

30

40

50

ELF

/ES

F (

%)

ASDNHS VSD TOF


MF band but its intensity is much weaker than the one at LF band. General auscultationof HS, usually includes respiratory sound, and the pulmonary sound components aremainly distributed at LF and MF bands.

Regarding the MF band, Figure 10 shows the plots of ICMMF for NHS and CHDs.The average and standard deviation of NHS’s ICMMF is 2.39±1.03 and the threshold ofthe cardiac murmurs at MF band is set as TCMMF = 3.42. It is found that there are 11ASD, four VSD and no TOF samples below the level of TCMMF indicating more thanhalf ASD patients are misdiagnosed. However, the misdiagnosed number for TOF hasbeen improved with the comparison of the results obtained at LF band.

−1

0

1

SF

−0.5

0

0.5

LF

−0.2

0

0.2

0.4

MF

1 2 3 4 5 6 7 8

−0.1

0

0.1

HF

Time (sec)

Figure 9. Example of pulmonary sound due to a normal breath, indicated by inFigure 8.

A

0 10 20 300

5

10

15

20

25



Sample number

EM

F/E

SF (

%)

0

5

10

15

20

25

EM

F/E

SF (

%)

0

5

10

15

20

25

EM

F/E

SF (

%)

ASD VSD TOFNHS

3.42B

Figure 10. Distributions of the indexes of cardiac murmurs at MF band (ICMMF).

Point in Figure 10 has the highest value of ICMMF in NHS. Its time waveformsreconstructed at four expected frequency bands are plotted in Figure 11. It is clear thatthe breath sounds represent the strong components in LF, MF and HF bands. The energydistributions of the breath component have almost the same intensity at both LF and MFfrequency bands.

The plots of ICMHF are shown in Figure 12. The average and standard deviation ofICMHF for NHS is 0.38±0.16 and the corresponding threhold of the cardiac murmurs isgiven as TCMHF = 0.54. The numbers of CHD samples below the threhold TCMHF are two

B


−1

0

1

SF

−0.5

0

0.5

LF

−0.5

0

0.5

MF

1 2 3 4 5 6 7 8

−0.2

0

0.2

HF

Time (sec)

Figure 11. Example of pulmonary sound due to a normal breath, indicated by inFigure 10.

B

0.541

0 10(i) (iii)

(iv)

20(ii) 300

1

2

3

4

5



Sample number

EH

F/E

SF (

%)

0

1

2

3

4

5

EH

F/E

SF (

%)

0

1

2

3

4

5

EH

F/E

SF (

%)

ASDNHS VSD TOF

Figure 12. Distributions of the indexes of cardiac murmurs at HF band (ICMHF).

in ASD(i,ii) (▼), two in VSD(iii,iv) (*) and none in TOF samples shown in Figure 12.The misdiagnosed cases are reduced significantly. This indicates that the cardiac murmursevaluation index at HF band ICMHF can be an evaluable index for quantitative diagnosisof HS murmurs. It is also evident that the diagnosis accuracy could be improved if the HSis recorded by a momentary stop of the breath, and in a quiet environment.

Based on the experimental investigation, the evaluation indexes of cardiac murmurs(ICMLF, ICMMF and ICMHF) are discussed. Performance measures such as TruePositives (TP), False Negatives (FN), True Negatives (TN), and False Positives (FP),accuracy, sensitivity (Sn) and specificity (Sp) are evaluated. TN is the number ofnormal cases correctly identified as normal. FN is the number of CHD cases incorrectlyidentified as normal. TP is the number of CHD cases correctly identified as they are,and FP is the number of normal cases incorrectly identified as CHD. Sensitivity(TP/(TP+FN)) is the ratio of CHD samples that are correctly identified to the totalCHD. Specificity (TN/(TN+FP)) is the ratio of normal samples are correctly identifiedto total normal HS. Accuracy (TP+TN)/(TP+FN+TN+FP) which is the ratio of thenumber of correctly identified samples to total number of samples [25]. Theperformance measures for the evaluation indexes of cardiac murmurs are shown inTable 3. It can be seen that performance measures of ICMHF yielded the highestsensitivity of 93.8% (the correct rate of CHD cases), specificity of 100% (the correctrate of normal cases) and accuracy of 95.3% (the correct rate of total cases),respectively, significantly better than those with ICMMF and ICMLF.

As mentioned above, the energy distribution at VLF band is considered to be avibration or systaltic movement in the body due to the heartbeat shown in Figure 7. InFigure 12, the four samples below the threshold of TCMHF = 0.54 might bemisdiagnosed as normal samples. Further analysis on these 4 CHD samples andcomparing with NHS revealed that the energy ratio EVLF of ASD(i), VSD(iii, iv) areover 20%, which are higher than that of NHS (see Table 4). This means ASD(i) andVSD(iii,iv) can be distinguished from NHS by EVLF (%).

Therefore, only the energy distribution of ASD(ii) is close to NHS. However, basedon the records of ASD(ii), the patient was diagnosed as a single ostium primum atrialseptal defect without mitral regurgitation or left ventricle to right atrium shunting, whilethe other ASD cases in this study are ostium secundum atrial septal defects. Patientswith smaller ostium primum atrial septal defects (little or no mitral regurgitation or leftventricle to right atrium shunting) are usually asymptomatic. Considering the analysisresult of evaluation indexes ICMHF and the four special cases of CHD (ASD(i,ii) and


Table 3. TN, FN, TP, FP, sensitivity (Sn), specificity (Sp) and accuracy (Acc) ofevaluation indexes of cardiac murmurs ICML

ICML TN FN TP FP Sn (%) Sp (%) Acc (%)

ICMLF 20 16 48 1 75 95.2 80ICMMF 20 15 49 1 76.6 95.2 81.2ICMHF 21 4 60 0 93.8 100 95.3

VSD(iii, iv)) at VLF band, we can attain improved performance measures withsensitivity of 98.4%, specificity of 100% and accuracy of 98.8%, respectively.

5. CONCLUSIONCHDs are abnormalities in the heart’s structure present at birth, which can range frommild to severe. Heart murmurs are pathological sounds produced by turbulent bloodflow due to cardiac defects. This paper has described a new approach on analysis ofthe pathologic cardiac murmurs based on the WPD technique which analyses bothlow and high frequency sub-bands of HS signals, collected from clinical subjects ingeneral environment. The HS signal was divided into five bands and the energyintensity at each frequency band was calculated and compared. Based on the analysisof clinic HS data, three evaluation indexes for cardiac murmurs were proposed for theanalysis of the pathologic murmurs. Finally, the threshold values between theinnocent and pathologic murmurs were determined based on the statistical results ofthe normal HSs. A case study on the NHS and CHD signals was performed to validatethe usefulness and performance of the proposed method. The performance measuresof ICMHF yielded the highest sensitivity of 93.8%, specificity of 100% and accuracyof 95.3%, respectively. Furthermore, considering the analysis result of evaluationindexes ICMHF at HF (344-1378Hz) and the four special cases at VLF band (5-21Hz),we can obtain the improved performance measures with a sensitivity of 98.4%,specificity of 100% and accuracy of 98.8%, respectively.

ACKNOWLEDGEMENTSThis research was partially supported by the Japanese Ministry of Education, Culture,Sports, Science and Technology, Grant-in-Aid for scientific Research (C), 2009~2011,21560243; 2012~2014, 24560261 and The Sichuan Provincial Key Technology R&D Program(No. 2012GZ0019), China. We are grateful to Dr. Jinbao Zhang,Cardiothoracic Surgery, Chengdu Military General Hospital of People’s LiberationArmy, for his technical support and expert knowledge.

CONFLICT OF INTERESTWe confirm that there are no known conflicts of interest associated with this research.


Table 4. Energy ratios values at different frequency bands of CHD heart soundbelow the threhold TCMHF.

Sample No. EVLF (%) ESF (%) ELF (%) EMF (%) EHF (%)

NHS (Eave±Estd) 15.67±4.45 76.16±4.33 6.05±1.66 1.82±0.792 0.29±0.12ASD(i) 22.79 72.23 4.10 0.76 0.11ASD(ii) 9.48 83.72 5.59 0.98 0.22VSD(iii) 50.21 44.81 3.84 0.88 0.23VSD(iv) 38.93 54.75 4.84 1.20 0.27

REFERENCES[1] National Heart, Lung, and blood Institute, 2011. http://www.nhlbi.nih.gov/health/health-

topics/topics/chd/. Accessed June 14, 2014.

[2] Valafar H., Valafar F., Darvill A., Albersheim P., Kutlar A., Woods K.F., Hardin J., Predicting theeffectiveness of hydroxyurea in individual sickle cell anemia patients, Artificial Intelligence inMedicine, 2000, 18(2): 133–148.

[3] CONGENITAL HEART DISEASE, 2011. http://www.pted.org/?id=list#1. Accessed July 28, 2014.

[4] Amer Abdullah Lardhi, Prevalence and clinical significance of heart murmurs detected in routineneonatal examination, Journal of the Saudi Heart Association, 2010, 22(1): 25–27.

[5] Reichlin S., Dieterle T., Camli C., Leimenstoll B., Schoenenberger A. and Martina B., Initial clinicalevaluation of cardiac systolic murmurs in the ED by noncardiologists, American Journal of EmergencyMedicine, 2004, 71–75.

[6] Higuchi k., Sato K., Makuuch H., Furuse A., Takamoto S. and Takeda H., Automated diagnosis ofheart disease in patients with heart murmurs: application of a neural network technique, InternationalJournal of Medical Engineering and Technology, 2006, 30: 61–68.

[7] Sepideh Jabbari, Hassan Ghassemian, Modeling of heart systolic murmurs based on multivariatematching pursuit for diagnosis of valvular disorders, Computers in Biology and Medicine, 2011, 41(9):802–811.

[8] Saeid Sanei, Mansoureh Ghodsi, Hossein Hassani, An adaptive singular spectrum analysis approachto murmur detection from heart sounds, Medical Engineering & Physics, 2011, 33(3): 362–367.

[9] Samjin Choi, Youngkyun Shin, Hun-Kuk Park, Selection of wavelet packet measures for insufficiencymurmur identification, Expert Systems with Applications, 2011, 38(4): 4264–4271.

[10] Yuerong Chen, Shengyong Wang, Chia-Hsuan Shen, Fred K. Choy, Matrix decomposition basedfeature extraction for murmur classification, Medical Engineering & Physics, 2012, 34(6): 756–761.

[11] Zümray Dokur, Tamer Ölmez, Heart sound classification using wavelet transform and incrementalself-organizing map, Digital Signal Processing, 2008, 18(6): 951–959.

[12] Fatemeh Safara, Shyamala Doraisamy, Azreen Azman, Azrul Jantan, Asri Ranga Abdullah Ramaiah,Multi-level basis selection of wavelet packet decomposition tree for heart sound classification,Computers in Biology and Medicine 2013, 43 (10): 1407–1414.

[13] M. Hariharan a,*, C.Y. Fook a, R. Sindhu b, Abdul Hamid Adoma, Sazali Yaacoba, Objective evaluationof speech dysfluencies using wavelet packet transform with sample entropy, Digital Signal Processing2013, 23 (3): 952–959.

[14] Shivnarayan Patidar, Ram Bilas Pachori, Segmentation of cardiac sound signals by removing murmursusingconstrained tunable-Q wavelet transform, Biomedical Signal Processing and Control, 2013,41(4): 559–567.

[15] Sanjay R. Bhatikar, Curt DeGroff, Roop L. Mahajan, A classifier based on the artificial neural networkapproach for cardiologic auscultation in pediatrics, Artificial Intelligent in Medicine, 2005, 33:251–260.

[16] Abdulkadir Sengur, An expert system based on principal component analysis: artificial immunesystem and fuzzy k-NN for diagnosis of valvular heart diseases, Computers in Biology and Medicine,2008, 329–338.

[17] Sepideh Babaei and Amir Geranmayeh, Heart sound reproduction based on neural networkclassification of cardiac vavle disorders using wavelet transforms of PCG signals, Computers inBiology and Medicine, 2009, 8–15.

[18] Amir A. Sepehri, Joel Hancq, Thierry Dutoit, Arash Gharehbaghi, Armen Kocharian, A. Kiani,Computerized screening of children congenital heart diseases, Computer Methods and Programs inBiomedicine, 2008, 92(2): 186–192.

[19] Samit Ari, Koushik Hembram, Goutam Saha, Detection of cardiac abnormality from PCG signal usingLMS based least square SVM classifier, Expert Systems with Applications, 2010, 37(12): 8019–8026.


[20] Samjin Choi, Zhongwei Jiang, Cardiac sound murmurs classification with autoregressive spectralanalysis and multi-support vector machine technique, Computers in Biology and Medicine, 2010, 40(1): 8–20.

[21] Shuping Sun, Haibin Wang, Zhongwei Jiang, Yu Fang, Ting Tao, Segmentation-based heart soundfeature extraction combined with classifier models for a VSD diagnosis system, Expert Systems withApplications, 2014, 41(4): 1769–1780.

[22] Peter Libby MD, Robert O., Bonow MD, Douglas L., Mann MD FACC, Douglas P., Zipes MD,Braunwald’s Heart Disease: A Textbook of Cardiovascular Medicine, 2007, p:1583.

[23] Samjin Choi, Detection of valvular heart disorders using wavelet packet decomposition and supportvector machine, Expert System with Applications 2008, 35(4): 1679–1687.

[24] M. Nilsson, P. Funk, E.M. Olsson, B. vonSchele, N. Xiong, Clinical decision-support for diagnosingstress-related disorders by applying psychophysiological medical knowledge to an instance-basedlearning system, Artificial Intelligence in Medicine, 2006, 36: 159–176.

[25] Donna Giria, U. Rajendra Acharya b,c,*, Roshan Joy Martisb, S. Vinitha Sree d, Teik-Cheng Lima,Thajudin Ahamed VIe, Jasjit S. Surif,1, Automated diagnosis of Coronary Artery Disease affectedpatients using LDA, PCA, ICA and Discrete Wavelet Transform, Knowledge-Based Systems, 2013,37:274–282.

[26] T. Sawayama, Auscultation training by CD: Heart Sound, Publisher of Nankodo, 1994 (in Japan).

[27] K. Nakao, Online Bed Side Learning: Heart Sound Auscultation, available at, http://www.medic.mie-u.ac.jp/student/sinnzou.html (in Japan). Accessed June 14, 2014.


International Journal of

AerospaceEngineeringHindawi Publishing Corporationhttp://www.hindawi.com Volume 2014

RoboticsJournal of

Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014


Active and Passive Electronic Components

Control Scienceand Engineering

Journal of



RotatingMachinery


Hindawi Publishing Corporation http://www.hindawi.com

Journal ofEngineeringVolume 2014

Submit your manuscripts athttp://www.hindawi.com

VLSI Design



Shock and Vibration


Civil EngineeringAdvances in

Acoustics and VibrationAdvances in



Electrical and Computer Engineering

Journal of

Advances inOptoElectronics

Hindawi Publishing Corporation http://www.hindawi.com

Volume 2014

The Scientific World JournalHindawi Publishing Corporation http://www.hindawi.com Volume 2014

SensorsJournal of


Modelling & Simulation in EngineeringHindawi Publishing Corporation http://www.hindawi.com Volume 2014


Chemical EngineeringInternational Journal of Antennas and

Propagation




Navigation and Observation



DistributedSensor Networks


New Approach on Analysis of Pathologic Cardiac Murmurs...

Documents

Transcript of New Approach on Analysis of Pathologic Cardiac Murmurs...